Skip to content

AMOS 22: abdominal multi-organ segmentation, 600 CT and MRI with 15 organ labels

pl.load("amos22") loads this dataset. The AMOS benchmark (Ji et al., NeurIPS 2022): 500 CT and 100 MRI scans with 15 abdominal organs annotated; 240 training, 120 validation and 240 test images. Read case by case out of the 24 GB zip64 archive on Zenodo (HTTP Range: the 1 207-entry central directory costs 145 KB, one case its own size). Case ids below 500 are CT, 500 and above are MRI. ds.filter(split='Tr') or 'Va' keeps the labelled cases.

import pardesline as pl

ds = pl.load("amos22")
print(ds.describe())  # roles, frames, units, labels, licence, citation
s = ds[0]
obj = s.image  # fetched now, cached under ~/.cache/pardesline/datasets/amos22

Provenance and licence
Source remote archive read by HTTP Range: https://zenodo.org/records/7262581/files/amos22.zip
Scheme archive
Size of the whole source 23.7 GB
Licence CC-BY-SA-4.0
Commercial use allowed by the licence
Redistribution allowed
Licence note CC-BY-SA 4.0 as stated in the archive's dataset.json (checked 2026-10-13); the Zenodo record page lists its own licence field, TO CONFIRM that both agree before redistribution.
Terms https://zenodo.org/records/7262581
Homepage https://amos22.grand-challenge.org/
Cite Ji Y., Bai H., Ge C., et al. AMOS: A Large-Scale Abdominal Multi-Organ Benchmark for Versatile Medical Image Segmentation. NeurIPS 2022 Datasets and Benchmarks.
Samples grouped by id

Roles (what each sample contains)

Role Type Frame Units Labels / layout Pattern or file
image volume from the file header from the file header amos22/images{split}/{id}.nii.gz
Abdominal CT (ids < 500) or MRI (ids >= 500), NIfTI, LPS mm from the header
mask labels from the file header from the file header 1=spleen, 2=right_kidney, 3=left_kidney, 4=gall_bladder, 5=esophagus, 6=liver, 7=stomach, 8=aorta … (15 labels) amos22/labels{split}/{id}.nii.gz
15-organ label map (training and validation cases)

Standardized layout

Files are fetched on first access into the dataset cache and stay in their original format; ds.export(folder) writes every sample as standard files (PLY for points and meshes, MetaImage for volumes and label maps) with a .pardesline.json sidecar carrying frame, units, labels, licence and provenance, plus a manifest for the folder. Label maps become exact surfaces with to_mesh(label=k), DICOM series become one volume, raw arrays are decoded through the layout above.

ds = pl.load("amos22").take(2)
ds.export("./standardized")        # PLY / MetaImage + sidecars + manifest.json

Algorithms and models that run on it

Generated from src/pardesline/data/manifests/amos22.json on 2026-10-14.